person
),
Student(person
),
AdvisedBy(person
,
person
), and YearsInProgram(person,
integer
). Types include
person
,
publication
, and course
.
ai,
graphics, languages, systems,
and theory
and anonymized to protect the privacy of the individuals.learnwts -g
command (refer to the manual for more
details). Specify AdvisedBy as the query predicate using -ne AdvisedBy
(This will cause only the pseudo-likelihood of the non-evidence predicates to be optimized). Perform 5-fold cross validation, each time training
with all but 1 of the 5 areas, using the left out area for testing.
Note that multiple databases can be
specified using -t
ai.db,graphics.db
etc. ai, graphics, language,systems
}
to learn the weights of the MLN, and theory
as
the database whose values you will try to infer. learnwts -d -ne AdvisedBy
, again using AdvisedBy as your query predicate.learnstruct
command in Alchemy:-maxNumPredicates <int> | The maximum number of predicates allowed in any learned clause |
-maxVars <int> | The maximum number of variables in any learned clause |
-beamSize <int> | The size of the beam, e.g. how many candidate clauses to retain at each iteration. |
-minWt <double> | Candidate clauses must have a weight at least as large as this value or else are discarded. Thus, assigning a larger minWt will prune more candidates. |
-fractAtoms | If sampling ground atoms, the fraction of each predicate's ground atoms to draw. |
-penalty <double> | Each difference between the current and previous version of a candidate clause penalizes the (weighted) pseudo-log-likelihood by this amount. |
learnstruct
-maxNumPredicates 3 -maxVars 3 -noSampleClauses -fractAtoms 0.5 -minWt
0.1 -penalty 0.05 -i uw-empty.mln -o uw-ls.mln -t <dbfiles>
-withEM
option to learnwts
when using discriminative training. Allow for 40-60 minutes for training.MLN-Generative |
MLN-Discrminiative |
MLN-Your-Rules |
MLN-StructLearn |
MLN-VP-1 |
MLN-VP-10 |
MLN-VP-100 |